Alibaba's Page Agent is an open-source, JavaScript-based in-page GUI agent that controls web interfaces through natural language by reading the live DOM as text, rather than using screenshots or external browser automation tools like Playwright or Puppeteer. The agent lives inside the webpage as plain JavaScript, inheriting the user’s cookies, session, and authentication, and requires no separate backend. Its core technique, DOM dehydration, compresses the page into a FlatDomTree, stripping redundant markup so smaller text models can act precisely. The project is model-agnostic through any OpenAI-compatible endpoint, is TypeScript-first, and ships under the MIT license. The codebase builds on browser-use, from which its DOM processing and prompt are derived. Prompt-level safety and single-page scope are real limits, and server-side validation is recommended for risky actions. The best fit is for copilots and form-filling inside apps you own. The monorepo splits concerns into packages including @page-agent/core, page-agent, and @page-agent/page-controller, with operation allowlists to limit scope.